|
PureBytes Links
Trading Reference Links
|
Hi Stanislav,
Neurals can achieve quite allot. But there are limitations to what can be
expected. Just feeding a neural all kinds of data as input and expecting
to obtain a decent prediction as output is foolhardy.
A good starting point on your data inputs would be to attempt multivariate
regression analysis. This will give you a feel for what is available
within the data for prediction. Some other techniques exist also. Neurals
can be used to model non-linear relationships in the data quite well.
Basic statistical analysis will also show you how some inputs, i.e. Equity
Price, are increasing with no true mean from time frame to time frame.
Normalizing data will be a very important step. As you progress in
modeling, you will probably want to add chaotic modeling factors as well.
This gets pretty interesting but I have found difficult to obtain decent
results.
A few books that are worth acquiring, some of which are quite technical...
1. Neural, Novel & Hybrid algorithms for time series prediction, T. Masters
2. Time Series Prediction, Forecasting the future and understanding the
past, Edited By: Weigend & Gershenfeld
3. Numerical Recipes in C, Press, Teukolsky...
4. Time Series Analysis: Forecasting and control, Box & Jenkins
The first book above will give you a good feel for what is practical, the
second what is possible.
Hope this helps and best of returns
Aongus
-----Original Message-----
From: Stanislav Shokhor <shokhor@xxxxxxxxxxx>
To: metastock-list@xxxxxxxxxxxxx <metastock-list@xxxxxxxxxxxxx>
Date: Wednesday, February 04, 1998 9:15 AM
Subject: Unidentified subject!
>Hi, folks!
>Does anybody have an experience of working with neuronets? I mean if
>anybody does have a positive
>experience of forecasting stock proces using neuronets. I'll be glad to
>hear any ideas.
>
>Stanislav Shokhor.
>
|